Core Agent MCP

Remote execution layer for agentic systems. Expose 64 stable tools via Model Context Protocol — Hermes, OpenClaw, Claude Code, atau MCP client lain bisa pakai ini remote atau lokal.

Kenapa?

Agentic systems butuh remote execution layer yang reliable — bukan cuma shell access, tapi tools terstruktur, typed, dan terdefinisi yang return hasil konsisten. core-agent-mcp solusinya:

• 🛠 64 production tools — file, network, system, text, git, crypto, monitoring
• 🔌 Dual transport — stdio (local) atau HTTP SSE (remote)
• 🎯 Stable output — setiap tool return {success: bool, data: ..., error?: str}
• 🧩 Hermes-ready — drop-in via mcp_servers config

Quick Start

# Install
pip install core-agent-mcp

# Run stdio (local agent)
core-agent-mcp

# Run HTTP (remote agent)
core-agent-mcp --transport sse --host 0.0.0.0 --port 8931

Integrasi Hermes

Tambah ke ~/.hermes/config.yaml:

mcp_servers:
  core-agent:
    command: "core-agent-mcp"
    timeout: 120

Remote:

mcp_servers:
  core-agent:
    url: "http://your-server:8931/mcp"
    timeout: 120
    connect_timeout: 30

/reset sesi Hermes — semua tool muncul sebagai mcp_core_agent_*.

Integrasi OpenClaw / Claude Code

Claude Code (claude_code.json)

{
  "mcpServers": {
    "core-agent": { "command": "core-agent-mcp" }
  }
}

OpenClaw

mcp_servers:
  core-agent:
    command: "core-agent-mcp"

Tools (64 tools)

System (11)

Tool	Description
system_info	OS, kernel, hostname, CPU, Python
exec_command	Execute shell command (timeout, cwd, env)
get_env	Get env var value
list_env	List env vars (keys atau key=value)
which	Locate binary in PATH
disk_usage	Disk space per partition
memory_info	RAM + swap usage
cpu_load	Load avg + per-core usage
uptime	System uptime
list_users	User accounts (UID>=1000 + root)
hostname	System hostname

Filesystem (16)

Tool	Description
read_text	Read file (line range, max 1MB)
write_text	Write/overwrite file (auto mkdir)
append_text	Append to file
list_dir	Dir listing with metadata
file_info	File/dir metadata detail
find_files	Find by glob pattern
grep_files	Search content via regex
copy_path	Copy file/dir (recursive)
move_path	Move/rename file/dir
delete_path	Delete file/dir (safe: need recursive flag)
create_dir	mkdir -p
download_url	Download URL to file
checksum_file	MD5/SHA1/SHA256/SHA512
tree	Directory tree (depth-limited)
tail_file	Last N lines
head_file	First N lines

Network (7)

Tool	Description
http_request	Full HTTP (GET/POST/PUT/DELETE/PATCH)
dns_lookup	DNS A/AAAA/MX/NS/TXT/CNAME
tcp_check	TCP port open check
whois	Domain WHOIS lookup
ssl_cert	SSL certificate details
public_ip	Detect public IP
ping_host	ICMP ping (count-based)

Text & Data (20)

Tool	Description
grep	Regex search in text
count	Words/lines/chars/bytes
diff	Unified diff text A vs B
json_parse	Validate & parse JSON
json_format	Pretty-print JSON
base64_encode	Base64 encode
base64_decode	Base64 decode
url_encode	URL percent-encode
url_decode	URL percent-decode
hash	MD5/SHA1/SHA256/SHA512
uuid_v4	Random UUID v4
uuid_v7	Time-ordered UUID v7
timestamp	ISO/Unix/RFC2822 now
hex_encode	Text to hex
hex_decode	Hex to text
unixtime_to_iso	Unix sec → ISO 8601
iso_to_unixtime	ISO 8601 → Unix sec
csv_parse	CSV text → JSON array
json_to_yaml	JSON → YAML
yaml_to_json	YAML → JSON

Development (7)

Tool	Description
git_status	Working tree status
git_log	Commit log (N entries, filter author)
git_diff	Diff unstaged/staged/refs
git_branches	List local branches
git_commit_count	Count commits (branch/all)
git_show	Commit detail + diffstat
format_json	Pretty-print JSON

Meta (2)

Tool	Description
ping	Health check → pong + version
help	List all tools with descriptions

CLI

core-agent-mcp [OPTIONS]

Options:
  --transport TEXT    stdio (default) atau sse
  --host TEXT         HTTP host (default: 0.0.0.0)
  --port INTEGER      HTTP port (default: 8931)
  --timeout INTEGER   Tool timeout (default: 120s)
  --max-output INTEGER  Max output bytes (default: 51200)
  --workdir TEXT      Working directory
  --help

Env Variables

Variable	Default	Description
CORE_AGENT_TIMEOUT	120	Default tool timeout
CORE_AGENT_MAX_OUTPUT	51200	Max output bytes
CORE_AGENT_WORKDIR	$PWD	Working directory
CORE_AGENT_PATH_ALLOWLIST	-	Comma-separated allowed paths

Dev

git clone https://github.com/jefripunza/core-agent-mcp
cd core-agent-mcp
uv venv && source .venv/bin/activate
uv pip install -e .
core-agent-mcp

# Test via MCP inspector
npx @modelcontextprotocol/inspector core-agent-mcp

License

MIT